Load transfer: the thirty centimetres where transport robot projects actually fail

Autonomous navigation indoors and across industrial sites is no longer an open problem. A vehicle finds its way, avoids obstacles, holds a cadence and reports when something is blocking it. Anyone setting up a transport project today buys that capability rather than developing it.

What is new in every project, by contrast, is the transition between the load and the vehicle. It happens twice per order, it lasts a few seconds, and it is the only moment when the robot physically interacts with something that does not belong to it: a pallet, a roll cage, a tote, a rack, a dock.

That is exactly where the schedule stops. The driving route can be planned from a floor plan. The transfer can only be planned against the real load carrier, and the real load carrier is almost never what its manufacturer’s data sheet says it is.

Three inaccuracies meet at an automatic pick-up. First, the positioning accuracy of the vehicle — how precisely it comes to rest at the station. Second, the dimensional accuracy of the load carrier itself. Third, the position of that load carrier on or in the station: whether it sits skewed, was set down at an angle, or overhangs the edge.

Each of those three is normally discussed separately and warranted separately. They are rarely added together. Yet the arithmetic is simple and unforgiving: the sum of the three deviations has to stay smaller than the capture range of the pick-up mechanism. If it does not, the pick-up fails — not occasionally, but systematically on some proportion of movements.

The difference between a project at 99 per cent and one at 92 per cent successful pick-ups is operationally enormous, because every failed attempt calls a person to the station. That person is the one the robot was meant to relieve. How to read such success rates honestly rather than averaging them is set out in Reading robot success rates honestly.

The Euro pallet is held up as the model case for standardisation, and in a sense it is: 1200 by 800 by 144 millimetres, identical worldwide, and virtually every automatic conveyor system is calibrated to that dimension.

The figures next to it are rarely read. For the timber, EPAL states inherent tolerances of plus/minus 2 millimetres per element across the three dimensions of length, width and thickness; where drying takes place in a wood-fired kiln, those tolerances can reach 5 millimetres. That is the figure for a new pallet built to standard.

In circulation it looks different. Pallets swell, dry out, splinter, lose blocks and develop protruding nails. Under EPAL rules damaged pallets lose their exchangeability and present a safety risk, and repairs may only be carried out by licensed operations using the repair mark. For an automation project that means the condition of your pallet pool is not a peripheral logistics matter but a technical boundary condition of the pick-up.

This is the real error of reasoning in many projects. Existing load carriers — pallets, roll cages, mesh boxes, trolleys — evolved over decades for handling by people. And a person has a practically unlimited capture range.

A forklift driver corrects continuously. They see that the pallet sits three centimetres out of square, turn the wheel slightly, come at it again, lift at an angle and nudge it straight on setting down. They do this without noticing, hundreds of times a shift. The task has no tolerance for them because they are the tolerance.

A robot has exactly one attempt with exactly one capture range. It does not correct by feel but by what its sensors can reliably detect. So the same load carrier that was unremarkable through thirty years of manual operation can become the limiting factor once the operation is automated. The robot has not got worse. The silent correction has gone.

There are two fundamentally different designs for the transfer, and choosing between them is a cost question rather than a matter of taste.

The passive design leaves all the work to the vehicle. It drives under a trolley and lifts it, it picks up a pallet, it tows a train. The station is then a piece of floor with a marking, perhaps a stop. It has no controls, no power supply and no safety case of its own.

The powered design splits the work: the vehicle brings a roller or chain conveyor deck up to a fixed conveyor deck, and the two move the load across together. That is mechanically more elegant and more tolerant, because the load is transferred under guidance rather than picked up. It has just one consequence that regularly surprises people: the station is now itself a machine, with its own controls, its own power supply, its own conformity assessment, and an interface over which both sides have to agree when the transfer happens.

That second machine belongs in the specification and in the budget. If it first appears during commissioning it becomes the most expensive line in the project, because electrical design, acceptance and floor space then have to be arranged retrospectively.

Anyone running a mixed fleet meets the interface at this point. VDA 5050 knows load changes as actions: pick and drop, each with optional parameters such as the duration for which the action stays in the RUNNING state. The vehicle state carries a field for the loads currently held; that field is optional, and an empty field tells master control that the vehicle can reason about its own load state and is currently carrying nothing. A vehicle can pick up and carry only one load at a time and must drop it before picking up the next.

That is precise, and it is limited. What is standardised is the message: that a pick-up was requested, that it is running, that it has completed. What is not standardised is the mechanism: how high it lifts, with what capture range, against which stop, with what feedback on a failed pick-up. Two vehicles from different manufacturers can both report pick in full conformity and still fail differently at the same station.

Where else the interface stops carrying is set out in VDA 5050 and the mixed fleet. For the transfer, the short version is this: the standard tells you that something happened, not that it will fit.

Vehicle selection, navigation, the economics and the path from an existing tugger fleet to autonomous vehicles are covered at length elsewhere in our archive — among others in Replacing the tugger train and Building-to-building material transfer. What those deliberately leave open is this list.

Before any equipment selection, settle: which load carrier exactly, by type designation and not by category. What condition it is actually in across the estate, measured from a sample rather than estimated. Whether it is the same at both ends or whether the load is transhipped on the way. How high the transfer surface sits, at both ends. Whether the station may stay passive or has to be powered. And what should happen when a pick-up fails: retry, defer the order, call a person.

That last point matters most and is forgotten most often. A failed attempt is not an exceptional case but a recurring event with a frequency you should know before you build a shift plan on it. Settling the fallback behaviour in writing beforehand means not negotiating it later in live operation.